OSSINING, N.Y.— Three years ago, a computer named Watson won $1 million on national television when it beat two contestants on “Jeopardy!” by betting $17,000 on “Bram Stoker” in the game show’s final round. (Host Alex Trebek’s clue? “William Wilkinson’s ‘An account of the principalities of Wallachia and Moldavia’ inspired this author’s most famous novel.”)
Today, IBM is making its own wager. In January, the company funneled about $1 billion into a new business unit dedicated to selling the software behind Watson, rebranding it from a sci-fi novelty act to an analytics engine capable of transforming a multitude of established industries.
As part of that shift, Watson is headed to Washington: IBM is aggressively marketing the technology to the federal government, in addition to hospitals, retailers and other potential customers. Last month, IBM began offering Watson and other big data services to federal health-care organizations — whether to process health records, diagnose diseases or help researchers process medical literature, among other applications.
Watson’s success depends on more than just sales, IBM says — it also needs to inspire an ecosystem robust enough to coax businesses to embrace artificial intelligence and find new uses for the technology. IBM has been giving away Watson software to developers, encouraging them to build their own apps using cognitive computing. In April, for instance, IBM sponsored a competition at the University of Maryland where students were asked to develop business applications for Watson.
Still, Watson is a work in progress. As IBM researchers refine the software, the company is still ironing out the details of how to commercialize it, and the promise may be years away.
At the sprawling Thomas J. Watson Lab here in Ossining, one of three such labs in the United States, scientists are developing Watson’s trademark “cognitive computing” software, which can interpret speech and text, rapidly mine large volumes of data, answer questions and learn from its mistakes. “Watson” refers to any IBM software using this technology. Competitors, such as Intel, are working on similar software.
Watson no longer exists in a mainframe the size of a small room (though the original is kept in the lab basement as a relic). Today, the same processing power fits into a computer the size of three stacked pizza boxes, and much of the software can be accessed from the Internet cloud.
In Ossining, one group of scientists is building Watson’s medical applications, such as the Watson Engagement Advisor, which crunches patient records and medical literature to suggest treatment options to physicians. It presents options in order of confidence in the treatment’s success — say, 80 percent for chemotherapy, 70 percent for drug therapy, also taking into account a patient’s preference. It might ask a physician to perform follow-up tests before the software offers suggestions.
The system uses roughly the same technology that allowed Watson to churn through an encyclopedia of knowledge before buzzing in on “Jeopardy!”
After the Jeopardy! show aired in 2011, IBM started receiving calls from doctors interested in using some form of Watson to process medical data, Mike Rhodin, who heads the Watson group, said in an interivew. “Healthcare wasn’t the first [application] I thought of. What was interesting was the doctors actually saw something in it that intrigued them.”
Over the past couple of years, IBM has been hawking Watson in Washington. The company recently announced new investments in its federal health-care practice, marketing Watson health IT software to potential customers such as the Defense Department. The company also named its first chief medical information officer for the federal practice. Last year, several members of the Watson team demonstrated its medical applications to lawmakers on Capitol Hill, primarily to educate them about the system, according to the company.
Still in its early stages, Watson has yet to make a dent in the local market. IBM only has a handful of Watson pilots running — at the Memorial Sloan Kettering Cancer Center in New York, and the M.D. Anderson Cancer Treatment Center in Houston, for instance.
The region’s earliest adopters will likely be government agencies applying Watson in health care, Anne Altman, IBM’s general manager for federal and government industries said in an e-mail. The Department of Veterans Affairs and the Defense Department, for instance, have had electronic health records for about 20 years, but are today “inundated with data and in need of new approaches to help make fact-based decisions more quickly.”
Eventually, agencies could use Watson to connect directly with citizens, Altman suggested. “Think of the possibilities of using Watson to improve citizen satisfaction and create something like a ‘411 for government’ that helps people navigate through the maze of federal agencies to quickly find the best, most accurate answers.”
IBM is still searching for a viable business model for Watson’s health applications — some, such as Sloan Kettering, partner with IBM to develop Watson’s capabilities. Others pay a version of a licensing fee, but the pricing varies and IBM declined to share how much.
“As you can imagine, spending two years of IBM researchers’ time and experts in cancer at Sloan Kettering is a very expensive proposition,” said Pierre Darmon, IBM’s content manager of sales and distribution.
“There is a need to have a solution which is not necessarily as deep in knowledge and narrow in focus so that ... the system does not require as much training, is much more general in nature in terms of the scope and can more easily be replicated from customer to customer. And that’s what we are currently doing.”
Other applications — such as a Watson financial adviser that might guide a user through a mortgage payment — require fewer resources and training to build and are easier to deploy, Darmon said.
And while healthcare is a complicated place to start, the fact that Watson’s early customers are hospitals could validate the technology, Rhodin said. The complexity of medical data has forced IBM to evolve Watson very quickly, making its entry into other fields easier, he added.
IBM researchers aren’t the only ones building Watson applications.
The Watson Group also manages a $100 million venture fund intended to support tech companies using Watson software, which the company has made available to developers via the cloud. (For now, access is free, though IBM plans to take a cut once the apps are sold.)
Its first investment was in Denver-based WellTok, which used Watson to create a medical assistant for patients, reminding them of appointments or suggesting physical activity. The second investment was in Fluid, a start-up building a virtual shopping assistant.
IBM is also reaching students — in addition to the competition at the University Maryland, the company announced last week that it co-designed cognitive computing courses at Carnegie Mellon University, New York University, Rensselaer Polytechnic Institute, University of California Berkeley, University of Michigan, Ohio State University and the University of Texas, which will be available in the fall.
In some cases, Watson is still learning. IBM scientists are still teaching it to understand simple linguistic elements, such as a concept known as negation.
“The biggest thing is disease X has these symptoms, disease Y shares these symptoms, but does not have something else. So the negation becomes very, very important,” IBM research scientist Jennifer Chu-Carroll said.
Watson also stumbled when it was asked to interpret text or speech involving “not” — it might interpret “the patient did not have a fever at the hospital” as a patient having a fever, but not being at the hospital.
And though Watson might help oncology specialists at M.D. Anderson to identify treatment options for leukemia patients, it can’t substitute for the expertise of an experienced doctor, said Lynda Chin, chairman of the genomic department at M.D. Anderson.
“The experts have seen [patients] a thousand times,” she said, and sometimes their decisions are based on intuition that’s difficult to explain. “To begin to capture that intuition [or] experience ... for me that’s an aspirational goal.”
In the meantime, the novelty act hasn’t ended; IBM is still coming up with fresh marketing campaigns to get people interested in Watson. At the South by Southwest festival in Austin, for instance, chefs at an IBM food truck whipped up new dishes Watson had suggested — Austrian chocolate burritos or Creole shrimp lamb dumplings, for instance. And last week, IBM unveiled a prototype that can engage in debates; a demo argued both sides of the statement “the sale of violent video games to minors should be banned.”
But ultimately, Watson represents a more dynamic approach to computing that can keep up with the large amounts of data generated by analytics systems, IBM chief executive Ginni Rometty noted in a letter to shareholders last month. “[W]e need a new paradigm. These new systems are not programmed; rather, they learn, from the vast quantities of information they ingest, from their own experiences, and from their interactions with people.”
An earlier version of this article said that Memorial Sloan Kettering had bought a Watson system for a one-time fee. This version has been revised to reflect that Memorial Sloan Kettering is partnering with IBM to develop Watson.